Satellite Earth Observation for supporting biodiversity monitoring: a case study of ecosystem extent mapping in the Great Western Woodlands, Australia.

The Kunming-Montreal Global Biodiversity Framework (GBF) provides a roadmap for action on biodiversity loss with the definition of an ambitious set of goals to achieve sustainable development by 2050. This international agreement includes a monitoring framework comprising 43 headline indicators to track the effects of policy implementation on biodiversity. Many of these indicators require information on ecosystem structure, composition or functioning, some of which can be provided by satellite-based Earth observation (EO) data. Different EO sensors (e.g., optical, radar, LiDAR) can produce unique information on various ecosystem characteristics, and the large coverage and systematic periodicity of EO data facilitate tracking changes in indicators across different spatial and temporal scales.

As part of an initiative by the Committee on Earth Observation Satellites (CEOS) Biodiversity study team, this project focuses on the use of EO data for producing the GBF Headline Indicator A.2: extent of natural ecosystems. Our study area is the Great Western Woodlands (GWW), located in south-western Australia, a biodiversity hotspot as the largest temperate woodland ecosystem in the world. This region faces threats related to climate change impacts, particularly, increases in aridity conditions and in fire frequency. For these reasons, monitoring ecosystem extent in the GWW is essential for land management, and conservation efforts.

In this study, we evaluated the effects of optical and Synthetic Aperture Radar (SAR) data integration on ecosystem discrimination and assessed the performance of different machine learning algorithms in relation to classification accuracy. To achieve this, we used multi-source Analysis Ready Data on a cloud computing platform to produce a series of tests incorporating different input data and classification methods.

Preliminary results indicate that classification products including optical and SAR data have higher overall accuracy (91%) and improved discrimination between similar ecosystem types in the GWW region, compared to optical-only products (87%). Additionally, different machine learning algorithms resulted in classifications products with similar accuracy statistics, but large differences in feature identification and boundary definition between ecosystem classes. These results showcase how satellite EO data, as a consistent, cost effective and repeatable measurement, can support the production of biodiversity indicators for management and conservation purposes.